11 #ifndef EIGEN_SPARSE_QR_H 12 #define EIGEN_SPARSE_QR_H 16 template<
typename MatrixType,
typename OrderingType>
class SparseQR;
70 template<
typename _MatrixType,
typename _OrderingType>
75 using Base::m_isInitialized;
77 using Base::_solve_impl;
80 typedef typename MatrixType::Scalar
Scalar;
89 ColsAtCompileTime = MatrixType::ColsAtCompileTime,
90 MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
94 SparseQR () : m_analysisIsok(false), m_lastError(
""), m_useDefaultThreshold(true),m_isQSorted(false),m_isEtreeOk(false)
103 explicit SparseQR(
const MatrixType& mat) : m_analysisIsok(false), m_lastError(
""), m_useDefaultThreshold(true),m_isQSorted(false),m_isEtreeOk(false)
119 void analyzePattern(
const MatrixType& mat);
120 void factorize(
const MatrixType& mat);
143 const QRMatrixType&
matrixR()
const {
return m_R; }
151 eigen_assert(m_isInitialized &&
"The factorization should be called first, use compute()");
152 return m_nonzeropivots;
181 eigen_assert(m_isInitialized &&
"Decomposition is not initialized.");
182 return m_outputPerm_c;
191 template<
typename Rhs,
typename Dest>
194 eigen_assert(m_isInitialized &&
"The factorization should be called first, use compute()");
195 eigen_assert(this->rows() == B.rows() &&
"SparseQR::solve() : invalid number of rows in the right hand side matrix");
197 Index rank = this->rank();
200 typename Dest::PlainObject
y,
b;
201 y = this->matrixQ().transpose() * B;
205 y.resize((std::max<Index>)(cols(),y.rows()),y.cols());
206 y.topRows(rank) = this->matrixR().topLeftCorner(rank, rank).template triangularView<Upper>().solve(b.topRows(rank));
207 y.bottomRows(y.rows()-rank).
setZero();
210 if (m_perm_c.size()) dest = colsPermutation() * y.
topRows(cols());
224 m_useDefaultThreshold =
false;
225 m_threshold = threshold;
232 template<
typename Rhs>
235 eigen_assert(m_isInitialized &&
"The factorization should be called first, use compute()");
236 eigen_assert(this->rows() == B.rows() &&
"SparseQR::solve() : invalid number of rows in the right hand side matrix");
239 template<
typename Rhs>
242 eigen_assert(m_isInitialized &&
"The factorization should be called first, use compute()");
243 eigen_assert(this->rows() == B.
rows() &&
"SparseQR::solve() : invalid number of rows in the right hand side matrix");
257 eigen_assert(m_isInitialized &&
"Decomposition is not initialized.");
265 if(this->m_isQSorted)
return;
269 this->m_isQSorted =
true;
306 template <
typename MatrixType,
typename OrderingType>
309 eigen_assert(mat.isCompressed() &&
"SparseQR requires a sparse matrix in compressed mode. Call .makeCompressed() before passing it to SparseQR");
314 ord(matCpy, m_perm_c);
315 Index n = mat.cols();
316 Index m = mat.rows();
317 Index diagSize = (std::min)(m,n);
319 if (!m_perm_c.size())
322 m_perm_c.indices().setLinSpaced(n, 0,
StorageIndex(n-1));
326 m_outputPerm_c = m_perm_c.inverse();
331 m_Q.resize(m, diagSize);
334 m_R.reserve(2*mat.nonZeros());
335 m_Q.reserve(2*mat.nonZeros());
336 m_hcoeffs.resize(diagSize);
337 m_analysisIsok =
true;
347 template <
typename MatrixType,
typename OrderingType>
352 eigen_assert(m_analysisIsok &&
"analyzePattern() should be called before this step");
358 Index nzcolR, nzcolQ;
367 m_outputPerm_c = m_perm_c.inverse();
380 const StorageIndex *originalOuterIndices = mat.outerIndexPtr();
381 if(MatrixType::IsRowMajor)
383 originalOuterIndicesCpy = IndexVector::Map(m_pmat.outerIndexPtr(),n+1);
384 originalOuterIndices = originalOuterIndicesCpy.
data();
387 for (
int i = 0; i < n; i++)
389 Index p = m_perm_c.size() ? m_perm_c.indices()(i) : i;
390 m_pmat.outerIndexPtr()[p] = originalOuterIndices[i];
391 m_pmat.innerNonZeroPtr()[p] = originalOuterIndices[i+1] - originalOuterIndices[i];
399 if(m_useDefaultThreshold)
402 for (
int j = 0; j < n; j++) max2Norm = numext::maxi(max2Norm, m_pmat.col(j).norm());
409 m_pivotperm.setIdentity(n);
419 mark(nonzeroCol) =
col;
420 Qidx(0) = nonzeroCol;
421 nzcolR = 0; nzcolQ = 1;
422 bool found_diag = nonzeroCol>=m;
433 if(curIdx == nonzeroCol) found_diag =
true;
439 m_lastError =
"Empty row found during numerical factorization";
446 for (; mark(st) !=
col; st = m_etree(st))
454 Index nt = nzcolR-bi;
455 for(
Index i = 0; i < nt/2; i++)
std::swap(Ridx(bi+i), Ridx(nzcolR-i-1));
458 if(itp) tval(curIdx) = itp.value();
459 else tval(curIdx) =
Scalar(0);
462 if(curIdx > nonzeroCol && mark(curIdx) !=
col )
464 Qidx(nzcolQ) = curIdx;
471 for (
Index i = nzcolR-1; i >= 0; i--)
473 Index curIdx = Ridx(i);
479 tdot = m_Q.col(curIdx).dot(tval);
481 tdot *= m_hcoeffs(curIdx);
486 tval(itq.row()) -= itq.value() * tdot;
489 if(m_etree(Ridx(i)) == nonzeroCol)
506 if(nonzeroCol < diagSize)
514 for (
Index itq = 1; itq < nzcolQ; ++itq) sqrNorm +=
numext::abs2(tval(Qidx(itq)));
527 for (
Index itq = 1; itq < nzcolQ; ++itq)
528 tval(Qidx(itq)) /= (c0 - beta);
535 for (
Index i = nzcolR-1; i >= 0; i--)
537 Index curIdx = Ridx(i);
538 if(curIdx < nonzeroCol)
540 m_R.insertBackByOuterInnerUnordered(
col, curIdx) = tval(curIdx);
541 tval(curIdx) =
Scalar(0.);
545 if(nonzeroCol < diagSize &&
abs(beta) >= pivotThreshold)
547 m_R.insertBackByOuterInner(
col, nonzeroCol) = beta;
549 m_hcoeffs(nonzeroCol) = tau;
551 for (
Index itq = 0; itq < nzcolQ; ++itq)
553 Index iQ = Qidx(itq);
554 m_Q.insertBackByOuterInnerUnordered(nonzeroCol,iQ) = tval(iQ);
558 if(nonzeroCol<diagSize)
559 m_Q.startVec(nonzeroCol);
564 for (
Index j = nonzeroCol; j < n-1; j++)
565 std::swap(m_pivotperm.indices()(j), m_pivotperm.indices()[j+1]);
573 m_hcoeffs.tail(diagSize-nonzeroCol).setZero();
577 m_Q.makeCompressed();
579 m_R.makeCompressed();
582 m_nonzeropivots = nonzeroCol;
588 m_R = tempR * m_pivotperm;
591 m_outputPerm_c = m_outputPerm_c * m_pivotperm;
594 m_isInitialized =
true;
595 m_factorizationIsok =
true;
599 template <
typename SparseQRType,
typename Derived>
603 typedef typename SparseQRType::Scalar
Scalar;
606 m_qr(qr),m_other(other),m_transpose(transpose) {}
607 inline Index rows()
const {
return m_transpose ? m_qr.rows() : m_qr.cols(); }
611 template<
typename DesType>
614 Index m = m_qr.rows();
615 Index n = m_qr.cols();
616 Index diagSize = (std::min)(m,n);
620 eigen_assert(m_qr.m_Q.rows() == m_other.rows() &&
"Non conforming object sizes");
622 for(
Index j = 0; j < res.cols(); j++){
623 for (
Index k = 0; k < diagSize; k++)
625 Scalar tau = Scalar(0);
626 tau = m_qr.m_Q.col(k).dot(res.col(j));
627 if(tau==Scalar(0))
continue;
628 tau = tau * m_qr.m_hcoeffs(k);
629 res.col(j) -= tau * m_qr.m_Q.col(k);
635 eigen_assert(m_qr.m_Q.rows() == m_other.rows() &&
"Non conforming object sizes");
637 for(
Index j = 0; j < res.cols(); j++)
639 for (
Index k = diagSize-1; k >=0; k--)
641 Scalar tau = Scalar(0);
642 tau = m_qr.m_Q.col(k).dot(res.col(j));
643 if(tau==Scalar(0))
continue;
644 tau = tau * m_qr.m_hcoeffs(k);
645 res.col(j) -= tau * m_qr.m_Q.col(k);
656 template<
typename SparseQRType>
659 typedef typename SparseQRType::Scalar
Scalar;
666 template<
typename Derived>
676 inline Index cols()
const {
return (std::min)(m_qr.rows(),m_qr.cols()); }
685 template<
typename SparseQRType>
689 template<
typename Derived>
699 template<
typename SparseQRType>
707 template<
typename DstXprType,
typename SparseQRType>
711 typedef typename DstXprType::Scalar
Scalar;
715 typename DstXprType::PlainObject idMat(src.
m_qr.rows(), src.
m_qr.rows());
718 const_cast<SparseQRType *
>(&src.
m_qr)->_sort_matrix_Q();
723 template<
typename DstXprType,
typename SparseQRType>
727 typedef typename DstXprType::Scalar
Scalar;
731 dst = src.
m_qr.matrixQ() * DstXprType::Identity(src.
m_qr.rows(), src.
m_qr.rows());
_OrderingType OrderingType
MatrixType::StorageIndex StorageIndex
PermutationMatrix< Dynamic, Dynamic, StorageIndex > PermutationType
Matrix< Scalar, Dynamic, 1 > ScalarVector
static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op< Scalar, Scalar > &)
EIGEN_DEVICE_FUNC Derived & setZero(Index size)
const AutoDiffScalar< DerType > & conj(const AutoDiffScalar< DerType > &x)
EIGEN_DEVICE_FUNC RealReturnType real() const
Derived::PlainObject ReturnType
int coletree(const MatrixType &mat, IndexVector &parent, IndexVector &firstRowElt, typename MatrixType::StorageIndex *perm=0)
const SparseQRType & m_qr
SparseQRType::MatrixType ReturnType
SparseQRMatrixQReturnType(const SparseQRType &qr)
Matrix< StorageIndex, Dynamic, 1 > IndexVector
const SparseQRType & m_qr
A base class for sparse solvers.
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar * data() const
storage_kind_to_evaluator_kind< typename MatrixType::StorageKind >::Kind Kind
SparseQR_QProduct(const SparseQRType &qr, const Derived &other, bool transpose)
EIGEN_DEVICE_FUNC const SqrtReturnType sqrt() const
Holds information about the various numeric (i.e. scalar) types allowed by Eigen. ...
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const AbsReturnType abs() const
Eigen::Index Index
The interface type of indices.
SparseQRMatrixQReturnType< SparseQRType > SrcXprType
SparseQRType::Scalar Scalar
EIGEN_DEVICE_FUNC RowsBlockXpr topRows(Index n)
ComputationInfo info() const
Reports whether previous computation was successful.
const QRMatrixType & matrixR() const
DstXprType::Scalar Scalar
SparseQRType::QRMatrixType MatrixType
void factorize(const MatrixType &mat)
Performs the numerical QR factorization of the input matrix.
SparseQR_QProduct< SparseQRType, Derived > operator*(const MatrixBase< Derived > &other)
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half() max(const half &a, const half &b)
static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op< Scalar, Scalar > &)
SparseQRType::MatrixType MatrixType
Sparse left-looking rank-revealing QR factorization.
IndexVector m_firstRowElt
EIGEN_DEVICE_FUNC ColXpr col(Index i)
This is the const version of col().
Base class of any sparse matrices or sparse expressions.
SparseSolverBase< SparseQR< _MatrixType, _OrderingType > > Base
EIGEN_DEFAULT_DENSE_INDEX_TYPE Index
The Index type as used for the API.
Matrix< Scalar, Dynamic, Dynamic > DenseMatrix
Base::InnerIterator InnerIterator
PermutationType m_pivotperm
DstXprType::Scalar Scalar
SparseQRMatrixQTransposeReturnType< SparseQRType > adjoint() const
EIGEN_DEVICE_FUNC Derived & setConstant(Index size, const Scalar &val)
const Solve< SparseQR, Rhs > solve(const SparseMatrixBase< Rhs > &B) const
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Abs2ReturnType abs2() const
MatrixType::Scalar Scalar
void evalTo(DesType &res) const
bool _solve_impl(const MatrixBase< Rhs > &B, MatrixBase< Dest > &dest) const
void analyzePattern(const MatrixType &mat)
Preprocessing step of a QR factorization.
SparseQRType::Scalar Scalar
PermutationType m_outputPerm_c
SparseQRMatrixQReturnType< SparseQRType > SrcXprType
SparseQRType::MatrixType ReturnType
const Derived & derived() const
const PermutationType & colsPermutation() const
const Solve< SparseQR, Rhs > solve(const MatrixBase< Rhs > &B) const
DstXprType::StorageIndex StorageIndex
SparseQRMatrixQReturnType< SparseQR > matrixQ() const
const SparseQRType & m_qr
EIGEN_DEVICE_FUNC const ImagReturnType imag() const
SparseQRMatrixQTransposeReturnType(const SparseQRType &qr)
ReturnType::StorageIndex StorageIndex
Pseudo expression representing a solving operation.
MatrixType::RealScalar RealScalar
ReturnType::StorageKind StorageKind
EIGEN_DEVICE_FUNC const Scalar & b
SparseQR(const MatrixType &mat)
Base class for all dense matrices, vectors, and expressions.
void swap(mpfr::mpreal &x, mpfr::mpreal &y)
SparseQRMatrixQTransposeReturnType< SparseQRType > transpose() const
DstXprType::StorageIndex StorageIndex
void setPivotThreshold(const RealScalar &threshold)
std::string lastErrorMessage() const
bool m_useDefaultThreshold
SparseMatrix< Scalar, ColMajor, StorageIndex > QRMatrixType
SparseQR_QProduct< SparseQRType, Derived > operator*(const MatrixBase< Derived > &other)
void compute(const MatrixType &mat)